Intelligent diagnostic optics for flow visualization

Optics & Laser Technology 32 (2000) 641–654

www.elsevier.com/locate/optlastec

Intelligent diagnostic optics for ow visualization P. Bryanston-Cross ∗ , M. Burnett, B. Timmerman, W.K. Lee, P. Dunkley Optical Engineering Laboratory, School of Engineering, Warwick University, Coventry, UK

Abstract A review of several current optical diagnostics used for ow visualization is presented and the limitations and strengths of each technique are described. A new type of intelligent diagnostic optic designed for making three-dimensional measurement of velocity in a gas turbine combustor is proposed. The diagnostic uses an in-line tomographic approach combined with correlation theory to spatially locate structure within the ow. A discussion is then made as to why some optical diagnostics have been more successful than others in their general application. The potential advantages of evolving new technology and the implications for future instrumentation are also c 2001 Elsevier Science Ltd. All rights reserved. discussed. Keywords: Tomographic optical ow diagnostics

1. Introduction For many years researchers have been investigating methods of quantitatively visualizing ows. The principal objective of this paper is to use the experience of the author, coupled with knowledge of evolving technology, to consider what we may expect from the next generation of intelligent diagnostic optics. The term ‘intelligent’ has been used to describe how computers can both be used to retain ‘knowledge’ and be able to use synthetic intelligence to make decisions on behalf of the operator. One way of looking into the future is to rst review the past. Optical instrumentation has moved from being a photographic static image-based approach to becoming a real-time active diagnostic with a wide range of applications. Over the past two decades a generation of optical diagnostics has been created and applied. With hindsight it is possible to understand why some methods have prospered and grown into accepted laboratory tools while others have remained bespoke specialist curios. There are usually two reasons for this: either the technology is not available to sustain the technique or the method is too application dependent. It is also of interest to analyse what new knowledge is being created and what new technological advances in optics and computing may be expected in the future.

∗

Corresponding author. Fax: +44-024-76-418922. E-mail address: [email protected] (P. Bryanston-Cross).

Currently, the range of ow measurement techniques available to be used for uid ow analysis can be divided into three categories: • Intrusive, based on physical probes introduced in the ow. • Partially intrusive, based on ow markers injected into the ow. • Completely non-intrusive, based on the optical detection of light-scattering properties of the ow. In the rst category, pressure and temperature probes and hot-wire anemometers [1] are used to provide local, steady-state measurements. However, they have the fundamental limitations of interfering with the measured ow and creating local disturbances, especially at high speeds. Optical techniques such as laser two focus (L2F) [2], laser Doppler anemometry (LDA) [3], Doppler global velocimetry (DGV) [4] and particle image velocimetry (PIV) [5] can be generally regarded as non-intrusive, since the size and concentration of the particles introduced into the ow as measurement aids are too small to change the ow behaviour. L2F and LDA are single-point techniques and provide good statistics for the time-averaged data at that point. For this reason, the main drawback is that spatial turbulent structures and transient eects in the ow cannot be resolved. Also, data acquisition can prove a lengthy process, dicult for transient rigs with short run-time and sparse seeding. These drawbacks are overcome by PIV and DGV, which provide multi-point, instantaneous measurements. PIV

c 2001 Elsevier Science Ltd. All rights reserved. 0030-3992/01/$ - see front matter PII: S 0 0 3 0 - 3 9 9 2 ( 0 0 ) 0 0 0 9 4 - 3

642

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

combines whole- eld visualization with the instantaneous capture of the data and is a robust technique for industrial environments. It is a non-intrusive measurement method, based on the optical detection and recording of light scattered by particulate ow markers. The spatial displacement between two images of the same marker captured at two well-de ned moments in time relates to the local velocity of the ow. Examples of totally non-intrusive techniques are interferometry [6] and laser-induced uorescence (LIF) [7], which rely on the optical properties of the ow itself and are mostly used in combustion applications (However, for the latter a uorescent tracer is added.) Interferometric measurements are whole- eld and two-dimensional. However, they are very sensitive to any optical deformations of the light path, especially optical anisotropy and the movement or vibration of optical surfaces. LIF is a promising technique, with the potential to determine density, temperature and velocity from the uorescent emission spectrum of the uid molecules. It is mostly suited to ows in which the emission can be easily separated from the noise and its success is still dependent on the charge coupled device (CCD) camera and laser technological advances. Given the emerging technology it is interesting to consider what new or enhanced diagnostics might be possible. As we move toward more intelligent computers and optical diagnostic systems, the balance shifts from simply being able to make a measurement to that of making a detailed comparison with computational prediction. A further shift has occurred, in that it is expected that the technique should be user friendly. This accommodates the average time a user stays with a particular experimental system, which both in University and industry has fallen markedly. As a consequence of this limited period of use, if the knowledge and technical requirement associated with making a technique work are demanding or there is a speci c skill dependency, it can have a short lifetime. For example, one reason holography was not pursued as a test technique in the USA was that it was perceived to be highly operator sensitive and thus vulnerable to the more mobile American workforce. This also shows why some institutions have been able to produce a high calibre of results. In these cases it is often the stability, continuity and consistent funding of the organisation, which has kept a critical skills base intact. A second reason that some techniques are more ‘bespoke’ than others is the question of setting up. Initially, the quality and stability of lasers was a major issue, while in some cases optical alignment has been a problem. A major success of LDA is that it has been perceived as a ‘point and shoot technology’. There is a backlash to this in that often technological problems have been simpli ed in order to promote sales. Thirdly it is always a surprise to realise that many of the simpler visualisation approaches such as shadowgraphy, Schlieren or focused Schlieren are not attempted simply because fewer people know how to apply them.

There are also some problems, which have yet to be solved using optical techniques, in particular measurements within combustion. This is also the region where most manufacturers consider that greatest gains in eciency and economy are still to be made. Until recently, there was neither the technology nor the computing power to extract the multispectral, spatially complex, and temporally incoherent data associated with the combustion process. However, new aspects of technology have arisen which enable previously unsuccessful methods now to emerge as successful diagnostics. Of additional signi cance, is the emergence of highresolution, self-calibrating and synthetically intelligent optical diagnostic systems. Much of the experience and knowledge, which made previous diagnostics possible were in fact related to person-dependent skills. Such skills can now be ‘incorporated’ into the software of the instrument. 2. Review of current methods We will now look at the fundamentals of some optical diagnostic systems with a view to considering how they may evolve and what new knowledge may be generated. 2.1. Laser Doppler anemometry (LDA) LDA is a non-intrusive optical technique for measuring the velocity of a uid at a point. Since the rst applications in turbomachinery [8] it has matured into a well-established optical technique that is currently commonly used in industry. 2.1.1. Principle of operation LDA allows up to three components of velocity to be measured at any one time over a small volume, the size of which is determined by the intersection of two or more beams of coherent light. The interference of the intersecting beams generates a grid of light within the measurement volume through which

ow tracing particles can pass, as shown in Fig. 1. A typical measurement is made over an area of 200 m × 200 m for a beam width of 50 m. The intensity of the light scattered by the particle as it passes through the measurement volume is modulated according to the spacing (S) of the fringes and the component of velocity (Vp) of the particle normal to the plane of the fringes. The frequency of the pulses is created by the light scattered o the particle as it passes through the spatially interferometrically fringe modulated laser beam. Fig. 2 gives a schematic of a typical laser Doppler anemometer in the backscatter con guration. The anemometer consists of two sections, the transmitting and receiving systems. The former consists of a coherent laser source, the beam of which is split and launched into polarisation-maintaining single-mode bres that relay the beams to the launch head. Each of the emerging beams is collimated and focused within the measurement volume.

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

643

Fig. 1. Laser Doppler anemometry fringe model. The intensity of the scattered light produced as a particle passes through a series of interference fringes contained within a measurement volume de ned by the intersection of two coherent beams. S = spacing between the fringes; t = time of ight of the particle between a fringe pair; Vp = velocity of particle.

Fig. 2. Laser Doppler anemometer — backscatter con guration.

With a measurement volume of 50 m the spatial resolution of an LDA is potentially high. However, in practice this is limited by the time required to traverse and map large areas in the ow. This is due to the time needed for sucient particles to pass through the measurement volume to create a statistically signi cant measurement. The close proximity of surfaces, to within a millimetre of the centre of the measurement volume, can mask the Doppler signal. Each of the channels must intersect within the same measurement volume to resolve the three-dimensional vector of a single particle. Accuracies range from ±2 to ±0:1% for prolonged acquisition periods in time-invariant ows. Velocity ranges of the order of less than 1 m=s to hypersonic have been measured. In order to maintain a sucient data rate there needs to be an abundance of seeding material, but these rates inevitably suer in boundary layers or areas of recirculation. High refractive index gradients such as those found in ame fronts and boundary layers also present problems as do the use of curved or non-homogenous materials in the construction of the viewing window.

2.1.2. Current development of the instrument LDA is conceptually very simple and highly portable. It has gained a wide acceptance as a measurement technique. However, in terms of the spatial temporal envelope it only provides a point measurement which is a temporal average of the data. To create a two- or three-dimensional reconstruction measurement of the ow eld requires a time intensive scan of the region of interest. The paradox of the technique is that the whole ow eld can only be mapped in detail accurately on a time-averaged basis by moving the interrogating spot from place to place. But in doing this LDA can never provide an instantaneous mapping. Fig. 3 is an example of an LDA measurement of a transonic ow through the rst-stage blade row of a compressor within a gas turbine engine. The image was provided by the Rolls-Royce Strategic Research Centre [9]. Such detailed images often require long periods of expensive rig running. With a cost of $10 to $80=point, 400 points=plane and 20 planes=passage to map a commercial complex transonic rst-stage compressor ow is relatively inexpensive. The results shown are of a compressor rotating at 10,000

644

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

Fig. 3. Use of LDA to plot the axial velocity of the air passing through a rotating gas turbine compressor blade row. Measured relative Mach number at (a) 35% span and (b) 90% span. (This image is reproduced with the permission of Rolls-Royce plc Strategic Research Centre [9].)

revolutions; with a peripheral blade speed of 340 km=s and a

ow speed of 500 km=s. The particles being tracked through the rotor are of the order of 0:1 m. This type of measurement is used to validate numerical simulation, in an application demanding a velocity measurement accuracy of the order of 2%. The results obtained have made an important contribution to the design of the gas turbine engine [8]. Apart from measurement time and cost there are other less obvious problems associated with LDA techniques. Firstly, the level of expertise needed to provide an intelligent result is substantial. There are also hidden layers of technological complexity, the operation of the laser and the critical launch of the beam into the bres, which transmit the light through a lens into the ow, being just two examples. An a priori good understanding of the ow is also needed to set up the sampling window for the collection of the data. After the measurement, the interpretation of the statistic nature of the data, particularly in turbulent ows, is also at the cutting edge of the subject. A description of how the third component of velocity may be measured is given in [10]. A further more complex problem is what is meant by the turbulence level [11]. It can be dicult to dierentiate between a genuinely turbulent ow and one that has intermittent coherent structures. The technique can provide accurate average values of turbulence but cannot be used to map the instantaneous bursting process within the boundary layer of a low-speed boundary layer ow, as shown by Fig. 7. There is also the stamina needed by the researcher to overcome the technical demands of long experimental data acquisition periods and the processing of complex three-dimensional data.

The hard lesson seems to be that the closer the real problem is approached, the more expensive and dicult it is to obtain a worthwhile measurement. Finally, there are the questions of who uses the results and to what use are they put? Predictive codes have now become much more accurate and advanced over the past decade. In particular, the unsteady three-dimensional nature of a transonic ow is now being predicted by Denton [12] and therefore there is a real need for matching optical diagnostic data of high accuracy. There are, in addition, certain levels of intelligence and experience needed to understand the uid mechanics, interpret and reconcile the measurements and the predictions and then know how to use this information. However, despite this and the complexity and diculty of the instrumentation, the LDA approach has resulted in the designer being able to increase the eciency of the gas turbine engine by considerable amounts. 2.1.3. The next generation of LDA instrumentation In addressing the issue of the future evolution of this technique, we can discern the following. To-date the cost of this type of instrumentation has not fallen. Although laser diodes and bre optics make the LDA head a much neater package, the price level remains at that associated with a specialized scienti c instrument. What may be expected from current technology is that such an instrument be integrated with current control software. This could take the form of a self-calibrating optical head, which would give a three-dimensional display showing where it is within a de ned CAD generated three-dimensional

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

map. In brief, we would have an intelligent probe with a ‘self ’-positional awareness. With the use of laser diodes and low-weight diraction-limited optics it should be possible to create a computer-controlled stepper motor which carries a lightweight optical head, allowing the operator to

y from point to point through the measurement space in a pre-con gured sequence. The use of high-powered diodes in a pulsed mode of operation allows a further possibility. The particle could be rst detected prior to entering the measurement volume. Then as it transits through the probe volume it could be illuminated by a high-intensity laser pulse. This would make it possible to track smaller particles at higher speeds. 2.2. Laser two-focus anemometry (L2F) L2F will only be mentioned brie y in this paper. It is an optical point measurement system similar to LDA, but projects two 250 m spots into the ow eld. A single particle breaking both spots creates a time-of- ight temporal signal. The intensity of the spots allows it to ‘see’ smaller particles than with the LDA approach, but it is more vulnerable to turbulence in the ow. Greater precision allows improved spatial ltering of the signal to give a greater tolerance to surfaces adjacent to the probe volume. Schodl [13] also describes a three-dimensional two-colour L2F system that has been packaged into a small rotatable optical head connected to a laser and photo-multipliers via bre optics, thus enabling three components to be measured from a narrow viewing angle. Schodl’s work illustrates what can be achieved when a critical expertise is supported and pursued over a number of years. As a consequence the work can be seen as highly in uential in the development of L2F. 2.3. Particle image velocimetry (PIV) 2.3.1. Principle of operation PIV is a technique that has the potential to make an instantaneous velocity measurement of the whole ow eld. In concept the technique is simple. It reduces the instantaneous measurement of the ow to an imaging process. A pulse laser is used as a high-intensity (100 mJ), short duration (10 ns) light source, the laser beam being shaped to form a narrow two-dimensional sheet of light (200 m). If the laser produces two or more pulses, then images of a particle motion in the eld can be stored. Two-dimensional velocity measurements are made in a particle-laden ow, the particles being illuminated by a double-pulsed laser sheet. An imaging device records pictures of the double-exposed individual particles. Given a known pulse separation and a measured particle image separation, velocity values within the plane can be determined. A representation of a PIV system is given in Fig. 4. Many publications have been made during the development of PIV. Examples of seminal work in this eld can be

645

Fig. 4. A schematic representation of a PIV system.

attributed to Meynart [14] and Adrian [15]. Meynart demonstrated the strength of the method by mapping the particle distribution in a Raleigh–Benard water ow. Flows which are now being studied at the microscopic level using what is termed MicroPIV [16]. It has, however, taken PIV many years to reach its current level of application and there are still signi cant data reduction problems in its application. It is a technique which has had to wait for both camera and computer technology to evolve in order to become a useful measurement tool. It still badly needs the application of synthetic intelligence in order to make it into a ‘real’ time diagnostic. From the simple PIV combination shown in Fig. 4 there are now many variants. Three-dimensional velocity measurements can be made using holographically stored particles, stereo cameras, single-camera defocusing=diraction and single-camera aperture masks [17]. Stereo camera con gurations can also be used to study 3D ows, notably by the applications of the Schiemp ug condition [18]. 2.3.2. PIV and direct ow visualisation: the concept of ‘the comfortable test environment’ In the experimental work completed for the automotive industry by the OEL [19], the primary drive has been to simplify the operation of the system to allow designers to test their own models. As the concepts of ‘just in time’ and rapid prototyping have become established as manufacturing practice, then the speed and ease of concept to product designing has led to shorter design timescales. For an engine block, for example, the design cycle for building and testing using rapid prototyping methods is now 3 to 6 weeks. Thus, the engine coolant ow diagnostics should be suciently simple and rapid in application that the engineers themselves may apply it. Thus, it would seem right to provide a PIV=Visualisation approach which places the light sheet and camera under the total control of the test engineer. The camera and image processing needs to be of sucient quality in terms of resolution and speed to provide a clear real-time visualisation of the ow. Use could be made of a ‘new technology’ miniature digital camera and a bre-optic-based laser light sheet delivery created from high-resolution low-weight diraction based optics. The PIV system could provide computerised control, which

646

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

Fig. 6. The stator wake. The image shows the wake from a stator blade as a dark slow moving region of uid. There is a classic shed vortex structure which can also be seen in Fig. 5. Other turbulent variations in velocity can be seen. Fig. 5. Velocity contour map constructed from PIV measurement made at MIT. It shows the air ow as it passes through a three-dimensional rotor spinning at 10,000 revolution=s.

is simple and light enough to be mounted on the arm of a low-cost industrial robot. This would allow test engineers to

y through the transparent models, of their design, identifying and recording problem ow areas for modi cation. Detailed PIV measurements could then be made and evaluated directly with CFD calculations. All diagnostics techniques have advantages and drawbacks, which balance complexity against accuracy. The underlining diculty with PIV is reconciling accuracy and calibration. Aerodynamic measurements typically are to be made with an expected velocity resolution of 2%; achieving this over a large eld of view in two or three dimensions is demanding in its precision [20]. PIV is a technique that is highly dependent, at high speed, on the size of seeding particles used and the method used to launch them into the ow. Creating a suitably dense seeding concentration of particles, small enough to follow the ow, is a serious consideration. Typically, in air, if the particles are greater than 0:2 m in diameter, there is a signi cant particle lag. 2.3.3. Applications of PIV The number of applications of PIV has grown rapidly. In Figs. 5 and 6 are shown images from a series of experiments performed upon a spinning transonic rotor [21]. The technique used allowed the visualisation of the instantaneous transonic ow in the inter-blade passage of a rotating annular cascade at engine conditions. By means of photographic recording of ow tracers and specialised image processing, the instantaneous velocity over the whole eld was measured with estimated accuracy between 2 and 4%. The velocity contour map shown in Fig. 5 was created by using a Delaunay triangular grid [20].

At the time of making this image the Nd=YAG laser used could only be run at 10 Hz. The wake passing frequency in the transonic rotating ow is in the region of 5 kHz. Thus, although it is possible to see unsteady features in the image, they are only instantaneous snapshots, as shown in Fig. 6. A double- ring laser capable of a pulse repetition rate of 10 kHz is required to visualise the motion of the unsteady

ow as it passes through the passage. Both the laser and the high-speed image intensi ed digital camera now existing have made this measurement possible. Thus, it is now technically feasible to study the instantaneous unsteady ow measurement through a transonic blade row passage. An example of a low-speed turbulent burst is shown in Fig. 7 and described in detail in [22]. The information that can be extracted is shown in Fig. 8. In this case the image was processed using purpose-written Fourier lter software, which removed the high and low spatial frequencies. It shows details of the trailing edge wake at the rear of the turbine blade and a normal shock has also been visualised. These are the expected results but the lter has also exposed the existence of a passage vortex structure travelling through the blade row, created by an upstream stator blade. A PIV measurement would require the combination of a high repetition laser with a high-speed image intensi ed digital camera to be able to track the progress of such a structure as it travelled through the turbine row. It also clearly needs suitable automated image processing software to expose the details of the ow eld. Another example of state-of-the-art PIV is the ow around two adjacent rotating cylinders [23, 24], as shown in Fig. 9. For this novel ow con guration, the cylinders rotate in opposite directions with their outer surfaces in contact. Steady-state tests were carried out for the following rotational Reynolds numbers: Re = 75; 485; 966 and 1470. In the above sets of experimental tests the longest aspect of the work was the processing of the data. This required

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

647

Fig. 7. The image shows the bursting action of a vortex within the boundary layer of a slow moving air ow. In this image a lter has been applied to remove the boundary layer pro le, revelling the bursting action.

from the processed image. It took 2 years of research to design and formulate the data reduction programme. This both discriminated between the signal and noise, identi ed the particle cross-section and augmented the Delaunay grid software. The Delaunay grid allows the plotting of irregular or sparse elds, performing velocity calculations between grid point centres which thus maintains the accuracy of the calculation [20].

Fig. 8. Flow features visualized using an FFT ltering approach. From top to bottom these are: (i) the upstream stator wake; (ii) a weak transonic normal shock and (iii) the trailing edge wake.

‘bespoke’ software, which searched for individual particle images, found the centre of each particle and then calculated the distance between possible pairs [21]. There is still a lack of available software to extract large-scale image features

2.3.4. The future The future for PIV is very promising. There is still a great deal of development to be done and much exploitation to come. The advent of high repetition Nd=YAG lasers and image intensi er cameras which can carry multiple CCD cameras makes it possible to consider being able to follow the high-speed evolution of turbulent structures. For example, we envisage being able to track the wake shed from an upstream rotor. A time-series set of images created by Westerweel [25] shows the transition from laminar to turbulent ow and the relaminarisation of a boundary layer. Perhaps the most evolutionary aspect is the use of synthetic intelligence in the creation of its PIV processing software [26]. The images shown in Figs. 5 and 9 were initially treated by ‘hand’ and took days to process. Using fuzzy logic software it was found possible to process the data in a few minutes. As the memories of computers increase and digital intensi ed cameras gain in resolution it will be possible to track complex ow events in real time. However,

648

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

Fig. 9. Low-speed ow between a pair of counter-rotating rollers.

the ability to track ow in three dimensions is still problematic. While the use of stereo cameras appears at face value to be no more complex than that already achieved by LDA, meeting, for example, the Scheimp ug condition needs very careful camera calibration [16]. Very encouragingly an experiment recently performed at MIT [27] showed that an image intensi cation of 200 was possible. Thus instead of requiring pulses of 100 mJ; 0:5 mJ would be sucient. The image-enhanced digital camera used also had a framing rate of 1 MHz. 2.4. Interferometry The next techniques to be considered as optical diagnostics for ow visualisation are interferometry, holography and the use of tomography. 2.4.1. General Interferometry is a ow visualisation method that is non-intrusive and can give quantitative results concerning the density distribution in a compressible ow. No probes or seeding have to be placed in the ow and the method can produce simultaneous information over the entire eld. It is based on the retardation that a light ray experiences when crossing an in-homogenous refractive index (density) eld. However, whereas with classical interferometry there is a requirement to path-match the reference and object beams, with holographic interferometry this requirement is relaxed. Holography also allows the use of low-cost

optical components in the formation of interferometric images. The phase delay is proportional to a line integral of the density along the light path through the ow eld. Once the phase delay is known the average density of the ow can be determined. From the density distribution using the isentropic ow equations, the Mach number can be derived as shown in Fig. 10. The image presented is that of a transonic ow through a two-dimensional gas turbine cascade and can be compared to a PIV image as in Fig. 5. The entire data may be stored holographically, which allows the phase information to be post-processed. Holography has also been used to store three-dimensional ow features such as those of a transonic shock front. The image shown in Fig. 11 shows the shockwave formed as an air ow enters the rotating rst stage compressor fan of a gas turbine engine [28]. On the negative side, as ow calculations [29] and experimental testing has progressed from being two- to three-dimensional, the value of images, such as shown in Fig. 10 has become limited. Both interferometry and holography require both mechanical and laser (SLM, TEM00 ) stability for beam-splitting interferometers. This restraint is however, relaxed for beam-sharing arrangements where the optical beams share common optical components and paths. 2.4.2. Holographic tomography (HT) To overcome the problems posed by trying to visualise three-dimensional refractive index elds, Vest [30] proposed the use of holographic tomography (HT). This technique has

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

Fig. 10. Interferogram of a transonic ow through a 2-D turbine cascade.

649

HT combines multiple interferometric views through the

ow to create a three-dimensional phase and hence density map. However, the experimental system demands a wide angle of view and a complex experimental construction. This is coupled with a priori assumptions essential to reconstructing three-dimensional refractive index elds. However, if the complexities of the system were embedded into the software and hardware of the digital cameras used, making the system ‘intelligent and self-aware’, the technique may yet emerge from its current restricted laboratory use. Currently, as a technique it is not widely accepted and has not made the contribution to uid dynamics that, for example, LDA has obviously done. Attempts have been made to create in-line tomographic approaches, notably, amongst others by Fomin [34], where the direct relationship between refractive index bending and the projected speckle pattern has been used to extract 3D information about the uid ow. The technique, however, requires assumptive arguments about the turbulent structure of the ow to overcome the need for a wide viewing angle. Thus, HT at the moment is a technique that has yet to realise its potential as an optical diagnostic. 3. A correlation based optical in-line tomographic diagnostic for combustion

Fig. 11. Reconstruction of holographic image of the transonic ow within a compressor blade row.

In what follows, we postulate how the previously described techniques can be combined using current technology, to assemble a novel instrumental approach for combustion diagnostics. 3.1. In-line optical tomography for combustion diagnostics

Fig. 12. A tomographic reconstruction made from eight holographic views. The image shows the turbulent structure of a high-speed jet ow.

been explored successfully by Watt [31], Cha [32], Parker [33] and Timmerman [6]. It has resulted in some spectacular achievements an example of which is shown in Fig. 12.

Work in the 1970s by Jakeman and Pike [35] explored how correlation approaches could be used to track large-scale ow features. In this paper we now propose to examine how the three previous techniques described can be combined into the operation of a new type of instrument. This instrument has been designed to operate in a high temperature, high pressure combusting ow to provide quantitative results. This has been found very dicult to achieve using any of the other techniques discussed. The combustion area is also where the power-generation industry sees the greatest eciency and economy gains to be possible. The proposed method, ‘in-line optical tomography’, was initially developed to investigate vortex bursting within a supersonic boundary layer [36]. The study showed that because the shape of the bursting vortex changes very quickly, an optical correlation of the shape was only possible for a length scale of approximately one boundary layer thickness. In this case the boundary layer thickness was 5 mm, a size co-incidentally similar to the expected size of the turbulent combustion cells [37]. Essentially, it is proposed that by applying correlation theory to the incoherent or random shape

650

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

Fig. 13. Water- ow analogy image of the combustion process.

of a turbulent burning cell, it can be spatially located and tracked in the ow. 3.1.1. Optical ltering of the combustion ame It has been shown that in several important combustion

ows the existence of OH is a signi cant indicator of the eciency of the burning process. OH is formed in the intermolecular burning region [38]. Because of its short lifetime, OH only exists in the ‘skin’ of the burning cell forming a layer between the burnt and unburnt gas. The image shown (Fig. 13a) is that of a water ow simulation of the intermolecular mixing region between the air and fuel mixing process with a uorescent marker added to the ‘fuel’ ow. With the use of a weak acid in the ‘fuel’ ow and a weak alkali in the ‘air’ ow, the uorescent eect is suppressed and the total intensity of the emitted light from a uorescent marker [39] is restricted to the molecular mixing region. A simulation of this has been achieved in Fig. 13b by edge enhancing the image. OH emits a speci c ultraviolet light 380 nm. If the light emitted from the ow is ltered, it can be used in a similar manner to visualise the intermolecular mixing region of the ame. In this case the problem of tracking the combustion ame has been reduced optically to that of tracking the light intensity being emitted from a complex bubble image structure. It may also be the case that the intensity of the emitted OH-generated UV emission gives a direct indication of the heat release of the ame. 3.1.2. Use of the random temporal behaviour of turbulent burning structures The image shown in Fig. 14a is that of a 0.1 m diam◦ eter atmospheric 1000 C burning propane gas jet made using holographic interferometry [40, 41]. The work was

performed to investigate the burning cell size as a function of induced swirl. The cell size measured from the interferogram is typically 4 mm. There are several ow features of interest visualised in the interferometric image. This phase image of the ame shows that there is little refractive index bending present. An empirical model shown in Fig. 14b of the ame shows that although the density gradients are high, the amount of bending is small due to the size and random nature of the burning cells. The cells also change in size and structure with distance from the burner. Finally, there is a characteristic chaotic structure to the ame. 3.1.3. In-line tomographic imaging Provided the combustion eld can be described as being constituted from turbulent structures, or irregular shapes moving within the ow, it should be possible to retrieve their location in three dimensions by using their irregular structure. In principle, only two optical projections, are needed to achieve this, as shown in Fig. 15. Let us imagine the irregular structures consisting of an OH skin, which emits light at a speci c wavelength. This example is based on combustion ames, where OH exists only in the molecular mixing region of the ame. With these OH structures travelling in the eld, the signal that is emitted from a single point will consist of pulses of light, as the OH edges pass. The timing between two pulses will depend on the shape and speed of a speci c structure. The intensity of the pulse will be dependent on the intensity of the burning process. Using two detectors, at a slight angle to each other, a correlation between the two signals can be made, from which the velocity of the structure can be derived. This approach is similar to the velocity correlation applied to particles passing through the measurement volume in an LDA system.

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

651

Fig. 14. Interferometric measurements of a propane burner: (a) Propane gas ame, 0.1m diameter. (b) Empirically derived model of the combustor.

Fig. 15. The process of turbulent structures measurement and correlation using two-line detectors.

A two detector system that has a viewing line that crosses at a certain point, shown in Fig. 15. An OH structure passing through that point will send a signal to both detectors. Each detector will receive signals from all other points along its respective line-of-sight, but only for the overlapping viewing area (point) will they nd a strongly correlating signal. This correlation result will occur provided the eld is suciently incoherent. The time delay between the arrival at the two detectors is also shown in Fig. 15. In Fig. 15 the two plots show dierent signals. Particle (a) appears in channels (1) and (2) simultaneously. Particle (b) appears in channel (2) before it appears in channel (1), hence there is no correlation. Fig. 16a represents how a line array (1) and single-line (2) detector can be combined to monitor structures as they pass through the measurement volume. Using their output signals, the speci c correlation signal from each point is used to locate each structure along the line A. This is similar to the ideas used in optical coherence tomography, or time-of- ight measurements, where use is made of light sources with a short coherence length to determine the spatial location of a signal. This principle has been generalised in Fig. 16b to two linear array detectors (a) and (b) mak-

ing it possible, using the correlation approach, to map the two-dimensional plane shown. Thus, Fig. 16 represents an extension to the concepts presented in Fig. 15a and b. 3.1.4. Unresolved aspects of the in-line tomographic approach There are many questions with respect to this proposed approach, which have yet to be explored experimentally: • What resolution can be obtained from the signal? • What is the optimal angle between the sensors? • Is the mathematical description of the ‘bubble’ model valid? • How incoherent is the turbulent structure and what value of correlation can be achieved from this type of signal processing? • Over what length of time should the signal be sampled to achieve a signi cant correlation coecient? • How sensitive is this technique for ‘noise’=small signals? • What is the sensitivity of the detector with respect to its distance from the turbulent combustion source? Although intensity falls with at a 1=R2 relationship from the its source, the solid angle of area of light collection projected

652

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

Fig. 16. Detector structure and correlation for 2D and 3D elds: (a) Correlation measurement using a 3 sensor and single-line detector. (b) Area which correlates between two 4 sensor line detectors.

by the detector increases linearly with distance. Thus, it is expected that the sensitivity of the detector will fall with a 1=R relationship. • What percentage of the light emitted from the formation of OH in the intermolecular burning region is re-absorbed as it passes through other ame centres? The proposed instrument could be used to locate a volume spatially by forming a temporal cross-correlation product. If the individual turbulent structure of each cell gives it a local coherence, then, just as Fig. 15 shows a two-dimensional object being mapped onto a one-dimensional array, a three-dimensional image could then be mapped onto a two-dimensional array. Because the turbulent features in combustion are typically low resolution, a 1 mm scale would seem adequate for mapping onto two probes of 100 × 100 sensors. Eectively, low-resolution incoherent temporal turbulence data is being used to encode spatial position using two area detectors. Cross-correlation between these detectors localizes the position of the incoherent source. It may also be possible to relate the intensity of the correlation to local heat release from this structure. 3.1.5. Large-scale PIV So far the operation of this instrument is similar to that used to develop the initial correlators as used originally in LDA and L2F systems and reviewed at the start of this paper. At that time the technology limited what could be considered, namely a single-dimensional correlator. It is now possible using the area array devices to collect turbulent data for a structure in three dimensions. Once the essential structural mapping of the volume has been completed the data resembles that of large particles travelling within a uid. By using a combination of correlation theory and particle tracking the passage of the structure can be mapped in 3D as it travels through the uid. If, for example, a time delay corresponding to the transit time

Fig. 17. Particle velocity measurement by the use of a temporally delayed signal.

for particle (b) in Fig. 15 were added to channel (2) then a positive correlation would be achieved, as shown in Fig. 17. This is exactly how the PIV data in the previous examples shown in Figs. 5, 6, 7 and 9 were processed. Thus, there is the potential in using this approach to extract the 3D velocity eld within the combustion zone. Previous work [41] demonstrates that the short-lived semi-coherent signature of a turbulence structure can be velocity tracked in this manner.

4. Conclusions In the main the most eective method of visualisation, when it is possible, is the simplest. The most useful diagnostic tests in industry performed by the OEL have been achieved when the system is simple enough for the designer to be able to control the test and see the results directly. In the case of low-speed coolant ows through an automotive engine 90% of the diagnostic knowledge has been gained in this manner. PIV has often been performed as an afterthought for experimental comparison with CFD.

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654

Thus, a simple video camera mounted orthogonally to a bre-optics-projected laser light sheet can be used to visualise a large number of ows. Using the current technology of near photographic resolution, digital cameras assist the resolution and accuracy with which such measurements can be made. When the latter has the addition of a high-speed electronic imaging camera the system rapidly becomes a high-speed ow visualization tool. What has been lacking until recently has been the ability to provide substantial software for extracting information from the images. NASA’s pioneering work in this eld shows that it is possible by using fuzzy logic, neural networks and genetic algorithms to simplify images that would have previously been considered impenetrable. Finally, the creation of a exible joystick control could permit the user to ‘ y’ through the ow. The time between product design and manufacture in almost all industries has fallen signi cantly. For example, this has resulted in the automotive industry in the necessity of providing results within weeks. Also the rate at which sta move between positions has increased, certainly to less than a 3 year cycle. The next generation of optical diagnostics should therefore, be simple, well packaged and incorporate the programming of a maximum amount of synthetic intelligence. What seems to make a particular optical diagnostic technique successful is the user’s perception of simplicity of operation, portability and the direct and clear presentation of information. While the diagnostic may be highly technologically complex, progressively the requirement is to package this so that the complexity is hidden within the instrument. It is also now feasible to create low-cost, lightweight and diraction-limited high-quality optics. This, coupled with high-resolution compact digital cameras, could mean that the next generation of instruments will resemble the type of diagnostic tools now emerging for medical applications. The aero companies are using optical technology less in their aerodynamic investigations. As their understanding and numerical predictive ability has grown in con dence, the optical techniques have ‘migrated’ to more fundamental areas of research, such as combustion. As the likely further gains in eciency from aerodynamic improvement have decreased, the cost of optical diagnostics has made them less attractive. This argument favours direct visualisation methods such as PIV and LIF, which have a strong potential for development. In these cases one can foresee that the designer may be able to pre-programme their inspection sequence and the type of ow features they wish to follow at the same time as they commit the CAD model for rapid prototyping. It would seem likely, particularly with the now limited supply of holographic materials, that HT will move to become a wholly electronic imaging technique. Again using lightweight, low cost, directly phase-sensitive digital cameras, a narrow-angle tomographic approach such as described in this paper could now emerge to provide a 3-D image of the ow eld directly.

653

Finally, the authors have described the potential for a new type of optical diagnostic, termed ‘in-line tomographic imaging’ drawn on the experience and development of several previous systems and created speci cally for application to combustion ows. This new diagnostic has been designed as part of a Data Fusion programme [42], which merges several types of diagnostic data in order that a single goal, in this case the spatial measurement of heat release, can be achieved. It is clear that if the existing complexities of the optics, software, theory and data interpretation can all be ‘wrapped’ into the software and hardware, a new type of intelligent diagnostic method could be created which provides the operator with what they may perceive as a direct visualization system. It is hoped that from such a device the same level of eciency gains that have been seen in the aerodynamics of turbomachinery can now be achieved via the comprehensive quantitative visualisation of the combustion process. Acknowledgements The authors wish to thank the following for their participation, involvement, inspiration and funding of the Intersect, Multi-sensored Intelligent Engine programme of which this paper represents part: Dr. D. Harvey, C. Goy & Dr. J. Black, Rolls-Royce Plc.; K. Brundish, Dr. C. Wilson, C. Hurley & R. Marsh, DERA Pyestock; Dr. M. Hauge CORUS (British Steel); Dr. A. Starr & Dr. P. Hannah, Manchester University; Dr. G. Kelly of NPL; and the personal support of Prof. R. Brook of Sira. Also to Dr. E. Hines, Dr. D. Udrea, T. Wilson & P. Johnson of the School of Engineering for their support and help. The authors wish to thank Rolls-Royce, Corus, DERA & EPSRC for their funding and support of this research work. References [1] Warburton A. Acquisition and analysis of hot wire anemometer data from the combustion chamber of a motored single cylinder research engine. Inst Mech Eng 1983;C78:1–11. [2] Schodl R. A laser-2-focus (L2F) velocimeter for automatic ow vector measurements in the rotating components of turbomachines. J Fluids Engng-Trans ASME 1980;102(4):412–9. [3] Hunter Jr. WW, Humphreys Jr. WM, Meyers JF. Application of point and global laser velocimeter techniques to supersonic ows. 80th Supersonic Tunnel Association Meeting, Cologne, Germany, October 1993. [4] Meyers JF. Development of Doppler global velocimetry as a ow diagnostics tool. Meas Fluids Combust Systems Special Issue: Meas Sci Technol 1995;6:769–83. [5] Bryanston-Cross PJ, Chana KS. PIV measurements made in the stator trailing edge wake rotor region in an annular transonic cascade. Proceedings of SPIE 3172.90, Paper presented at the SPIE Conference Optical Technology in Thermal and Combustion Flow, SPIE vol. 3783, San Diego, 27 July–1 August 1997. p. 332–8. [6] Timmerman, BH. Holographic interferometric tomography for unsteady compressible ows. PhD dissertation, Delft University of

654

[7]

[8] [9]

[10] [11]

[12]

[13] [14] [15] [16] [17]

[18]

[19]

[20]

[21]

[22]

[23]

P. Bryanston-Cross et al. / Optics & Laser Technology 32 (2000) 641–654 Technology, EburonP&L (Publisher), 25 November 1997, ISBN: 90-5651-044-4. Locke R, Zaller M, Hicks Y, Anderson R. Non-intrusive laser-induced imaging for speciation and patternation in high pressure gas turbine combustors. Paper presented at the SPIE Conference Optical Technology in Thermal and Combustion Flow, Denver, 17–21 July 1999 Strazisar AJ, Powell JA. Laser anemometry measurements in a transonic axial ow compressor rotor. ASME J Engng Power 1981;103:430–7. Edmonds JD, Harvey D, Wiseall SS. Recent developments in the application of laser Doppler anemometry to compressor rigs. AGARD Conference Proceedings 598, Paper 16, Advanced Non-Intrusive Instrumentation for Propulsion Engines, Brussels, 20 –24 October 1997. Menon RK. Three component velocity measurements in the interblade region of a fan. TSI Incorporated, International Gas Turbine Conference, 87-GT-207, Anaheim, CA, June 1987. Hascher HG, Jari K, Novak M, Lee K, Schock H. Michigan State University; and Bonne M, Keller P. Chrysler Corporation, An evaluation of turbulent kinetic energy for the in-cylinder ow of a four-valve 3.5L SI engine using 3-D LDV measurements. SAE Paper 970793, International Congress & Exposition, Detroit, Michigan, 24 –27 February 1997. Denton JD, Dawes WN. CFD for turbomachinery design. Proceedings, Institution of Mechanical Engineers Conference on the Successful Exploitation of CFD in Turbomachinery Design, London, December 1997. Schodl R, Forster W. Multicolor beroptic laser-2-focus velocimeter For 3-dimensional ow-analysis. AIAA J 1991;29(8):1290–7. Meynart R. Equal velocity fringes in a Rayleigh-Benard ow by a speckle method. Appl Opt 1980;19:1385–6. Adrian RJ. Particle imaging techniques for experimental uid mechanics. Annu Rev Fluid Mech 1991;23:261–304. Meinhart CD, Wereley ST, Santiago JG. PIV measurements of a microchannel ow. Exp Fluids 1999;27:414–9. Chan V, Barnhart D, Garner C, Heilwell N, Coupland J. A uni ed system for holographic measurement in uid and solid mechanics: application of this system to ow in a volumetric ow measurement in an i.c.engine. Paper presented at the SPIE Conference Optical Technology in Thermal and Combustion Flow, Denver, SPIE vol. 3783, 17–21 July 1999. p. 110 –7. Westerweel J, Van Oord J. Stereoscopic PIV measurements in a turbulent boundary layer. in: Stanislas M, Kompenhans J, Westerweel J, editors. EUROPIV: a cooperative action to apply Particle Image Velocimetry to problems of industrial interest. Dordrecht: Kluwer, 2000. p. 459 –78 (to appear). Udrea DD, Bryanston-Cross PJ, Driver C, Calvert G. PIV measurements of the coolant ow within a transparent model engine block. Paper presented at the SPIE Conference European Symposium on Lasers and Optics in Manufacturing, Munich, 16 –20 June 1997. Udrea DD, Bryanston-Cross PJ, Funes-Gallanzi M. High accuracy techniques applied to the extraction of absolute position estimates in a 3DPIV system. Seventh International Symposium on Applications of Laser Techniques to Fluid Mechanics, Lisbon, Portugal, 11–14 July 1994. p. 35.7.1–7. Epstein AH, Bryanston-Cross PJ. Optical diagnostics in turbomachinery. Paper presented at, AGARD Conference 90th PEP Symposium on Advanced Non-Intrusive Instrumentation for Propulsion Engines, Brussels Belgium, 20 –24 October 1997. Burnett M, Lee WK, Bryanston-Cross PJ. The instantaneous determination of a transonic stator wake and turbulent burst in a low speed boundary layer in air using particle image velocimetry (PIV). ImechE 1998, C541=035=98. Hines EL, Udrea DD, Wilson T, Dunkley P, Burnett M, Bryanston-Cross PJ. The application of fuzzy logic and neural networks to the automatic determination 2 and 3 dimensional PIV

[24]

[25]

[26] [27]

[28] [29]

[30] [31] [32]

[33] [34]

[35] [36]

[37]

[38] [39] [40]

[41] [42]

velocity distributions. Paper presented at the SPIE Conference Optical Technology in Thermal and Combustion Flow, Denver, 17–21 July 1999. Tucker PG, Pan Z, Udrea D, Wilson T, Dunkley P, Bryanston-Cross P. Comparison of particle image velocimetry with computations for

ow induced by adjacent spinning cylinders. The Third International Workshop on PIV ’99-Santa Barbara, Hosted by The University of California, Santa Barbara, 16 –18 September 1999. Westerweel J, Draad AA. Measurement of temporal and spatial evolution of transitional pipe ow with PIV. In: Adrian RJ, et al., editors. Developments in laser techniques and uid mechanics. Berlin: Springer, 1997. Wernet M. Maturation of DPIV into a facility hardened technique for turbomachiery research. In this Millennial Issue. Burnett M, Creviston D, Kirk D, Bryanston-Cross P, Waitz I, Timmerman B. The application of ow visualisation at MIT to measure the ejection density mixing rate of a high speed combustor. Paper presented at the SPIE Conference Optical Technology in Thermal and Combustion Flow, Denver, 17–21 July 1999. Bryanston-Cross PJ. High speed ow visualisation. Prog Aerosp Sci 1986;23:85–l04. Dawes WN. Simulation of unsteady blade row interaction with CFD: applications. In: Blade row interference eects in axial turbomachinery stages. VKI Lecture Series 1998-02. Von Karman Institute for Fluid Mechanics, 1998. Vest CM. Holographic interferometry. New York: Wiley, 1979. Watt DW, Vest CM. Turbulent ow visualization by interferometric integral imaging and computed tomography. Exp Fluids 1990;8: 301–11. Zhao T, Cha S. Reconstruction of a ow eld by holographic interferometric tomography. Proceedings of SPIE 3172.90, Paper presented at the SPIE Conference Optical Technology in Thermal and Combustion Flow, San Diego, 27 July–1 August 1997. Parker SCJ, Holden CME, Bryanston-Cross PJ. Study of transonic

ows by quantitative holographic interferometry. Proc Inst Mech Eng, Part G: J Aerosp Engng 1994;208(2):91–7. N. Fomin, Statistical analysis of turbulence by single exposure speckle photography. Proceedings of SPIE 3172.90, Paper presented at the SPIE Conference Optical Technology in Thermal and Combustion Flow, San Diego, 27 July–1 August 1997. Jakeman E, Pike E, Pusey P. Nature 1976;263:215. Bryanston-Cross PJ, Camus J-J, Richards P. Dynamic correlation of a Schlieren image in a transonic air ow, photon correlation techniques in uid mechanics, Springer-Verlag Series in Optical Sciences, vol. 38. Berlin: Springer, 1983. p. 270 –5. Bryanston-Cross PJ, Edwards J, Squire L. Measurements in an unsteady two dimensional shock=boundary layer interaction. Presented at the IUTAM Unsteady Aerodynamics Conference, Jesus College, Cambridge, September l984. Tokumaru PT, Dimotakis PE. Image correlation velocimetry. Exp Fluids 1995;19(1):1–15. Bryanston-Cross PJ, The measurement of mixing in a combustion

ow model. Presented at the Association for High Speed Photography, Churchill College, Cambridge, 2–3 April 1992. Bryanston-Cross PJ, Towers DP. Quantitative holographic interferometry applied to combustion and compressible ow research. Presented at the Advances in Holography Conference, Imperial College UK, 4 –7 August 1992. Bryanston-Cross PJ, Gardner JW. Holographic visualisation of a combustion ame. Opt Lasers Engng 1988;9(2):85 –l00, IS:0143-8166. Bryanston-Cross P, Burnett M, Udrea DD, Marsh R, Timmerman BH, Starr A, Estabon J. INTErSECT: the application of data fusion to a multi sensored intelligent engine. Invited Paper Presented at: IEE Colloquium on Intelligent and Self-Validating Sensors, 21 June 1999-08-25, Wolfson College, Oxford.